Interesting insights into how users ascribe human-like attributes to AI from this article on “The 4 Degrees of Anthropomorphism of Generative AI” by Sarah Gibbons at NN/g. It categorizes these behaviors into four degrees, each reflecting a deeper level of emotional engagement and perceived AI intelligence: (1) Courtesy - basic politeness in prompts like “please” or “thank you”; (2) Reinforcement - encouragement through feedback such as “good job!”, hoping to influence the AI's future behavior; (3) Roleplay - employing prompt skeuomorphism (e.g., “act as a marketing expert…”); (4) Companionship - forming an empathetic bond with the AI.
The article makes observations about why people anthropomorphize AI. However, it could also be argued that a significant factor is the transference of the existing mental model for chat, i.e. that the most intuitive way to interact with a conversational interface is to converse with it as we have done with other humans since the advent of IM/SMS, especially in the absence of explicit user guidelines or UI cues suggesting otherwise.
Anthropomorphism is just one example of cognitive bias. Beyond the scope of the article, a broader topic that needs to be considered is how cognitive biases surface in these human-AI interactions, both those that negatively impact the experience but also those that are beneficially adaptive. Examples include:
Authority bias - accepting the opinion or conclusion of an LLM without critical examination.
Anchoring bias - overly relying on the first response to a prompt without exploring alternatives (contrast the more singular nature of an LLM response vs. the multiple of options presented from SERPs).
Halo effect - letting positive experiences with specific prompts lead to a belief that all outputs from the model are excellent, thus overlooking the significance of any negative aspects.
Fundamental attribution error - attributing a model's incorrect or unsatisfactory answer solely to its inherent limitations or flaws (internal attribution), without considering external factors like the clarity of the prompt.
https://lnkd.in/g5RK8uZw
#designforAI #ux
